List your company

Mathematical Reasoning on GSM8K and AIME24

46.9Accuracy (GSM8K)FluxAttn (FA-SSA)
39.41241.35643.345.244Apr 8, 2026

Evaluation results

MethodLinks
FluxAttn (FA-SSA)Backbone=Qwen3-8B2026.0446.942.3544.63
FluxAttn (FA-TA)Backbone=Llama-3.1-8B-Instruct2026.0446.938.342.6
TriangleMixBackbone=Llama-3.1-8B-Instruct2026.0446.337.0541.68
FluxAttn (FA-SSA)Backbone=Llama-3.1-8B-Instruct, Configuration=Variant 22026.0445.937.3541.63
FluxAttn (FA-SSA)Backbone=Qwen3-8B, Configuration=Variant 22026.0445.343.244.25
FluxAttn (FA-SSA)Backbone=Llama-3.1-8B-Instruct2026.0445.337.0541.18
FluxAttn (FA-TA)Backbone=Qwen3-4B2026.044540.3542.68
DuoAttentionBackbone=Llama-3.1-8B-Instruct2026.0444.433.6539.03
FluxAttn (FA-XA)Backbone=Llama-3.1-8B-Instruct2026.0444.433.6539.03
FluxAttn (FA-XA)Backbone=Qwen3-8B2026.0443.235.6539.43
FluxAttn (FA-TA)Backbone=Qwen3-8B2026.044339.0541.03
Llama-3.1-8B-InstructBackbone=Llama-3.1-8B-Instruct2026.0442.330.3536.33
FluxAttn (FA-XA)Backbone=Qwen3-4B2026.0442.240.3541.28
PruLongBackbone=Llama-3.1-8B-Instruct2026.0441.329.8535.58
DuoAttentionBackbone=Qwen3-8B2026.0441.235.6538.43
TriangleMixBackbone=Qwen3-8B2026.0441.244.1542.68
Qwen3-8BBackbone=Qwen3-8B2026.0440.632.3536.48
PruLongBackbone=Qwen3-8B2026.0440.432.3536.38
TriangleMixBackbone=Qwen3-4B2026.0440.337.2538.78
FluxAttn (FA-SSA)Backbone=Qwen3-4B2026.0440.337.0538.68
FluxAttn (FA-SSA)Backbone=Qwen3-4B, Configuration=Variant 22026.0439.937.2538.58
Qwen3-4BBackbone=Qwen3-4B2026.0439.730.3535.03
DuoAttentionBackbone=Qwen3-4B2026.0439.737.0538.38
PruLongBackbone=Qwen3-4B2026.0439.730.3535.03